arch/sparc/kernel/sys_sparc_64.c

   1 /* linux/arch/sparc64/kernel/sys_sparc.c
   2  *
   3  * This file contains various random system calls that
   4  * have a non-standard calling sequence on the Linux/sparc
   5  * platform.
   6  */
   7
   8 #include <linux/errno.h>
   9 #include <linux/types.h>
  10 #include <linux/sched/signal.h>
  11 #include <linux/sched/mm.h>
  12 #include <linux/sched/debug.h>
  13 #include <linux/fs.h>
  14 #include <linux/file.h>
  15 #include <linux/mm.h>
  16 #include <linux/sem.h>
  17 #include <linux/msg.h>
  18 #include <linux/shm.h>
  19 #include <linux/stat.h>
  20 #include <linux/mman.h>
  21 #include <linux/utsname.h>
  22 #include <linux/smp.h>
  23 #include <linux/slab.h>
  24 #include <linux/syscalls.h>
  25 #include <linux/ipc.h>
  26 #include <linux/personality.h>
  27 #include <linux/random.h>
  28 #include <linux/export.h>
  29 #include <linux/context_tracking.h>
  30
  31 #include <linux/uaccess.h>
  32 #include <asm/utrap.h>
  33 #include <asm/unistd.h>
  34
  35 #include "entry.h"
  36 #include "kernel.h"
  37 #include "systbls.h"
  38
  39 /* #define DEBUG_UNIMP_SYSCALL */
  40
  41 asmlinkage unsigned long sys_getpagesize(void)
  42 {
  43         return PAGE_SIZE;
  44 }
  45
  46 /* Does addr --> addr+len fall within 4GB of the VA-space hole or
  47  * overflow past the end of the 64-bit address space?
  48  */
  49 static inline int invalid_64bit_range(unsigned long addr, unsigned long len)
  50 {
  51         unsigned long va_exclude_start, va_exclude_end;
  52
  53         va_exclude_start = VA_EXCLUDE_START;
  54         va_exclude_end   = VA_EXCLUDE_END;
  55
  56         if (unlikely(len >= va_exclude_start))
  57                 return 1;
  58
  59         if (unlikely((addr + len) < addr))
  60                 return 1;
  61
  62         if (unlikely((addr >= va_exclude_start && addr < va_exclude_end) ||
  63                      ((addr + len) >= va_exclude_start &&
  64                       (addr + len) < va_exclude_end)))
  65                 return 1;
  66
  67         return 0;
  68 }
  69
  70 /* These functions differ from the default implementations in
  71  * mm/mmap.c in two ways:
  72  *
  73  * 1) For file backed MAP_SHARED mmap()'s we D-cache color align,
  74  *    for fixed such mappings we just validate what the user gave us.
  75  * 2) For 64-bit tasks we avoid mapping anything within 4GB of
  76  *    the spitfire/niagara VA-hole.
  77  */
  78
  79 static inline unsigned long COLOR_ALIGN(unsigned long addr,
  80                                          unsigned long pgoff)
  81 {
  82         unsigned long base = (addr+SHMLBA-1)&~(SHMLBA-1);
  83         unsigned long off = (pgoff<<PAGE_SHIFT) & (SHMLBA-1);
  84
  85         return base + off;
  86 }
  87
  88 unsigned long arch_get_unmapped_area(struct file *filp, unsigned long addr, unsigned long len, unsigned long pgoff, unsigned long flags)
  89 {
  90         struct mm_struct *mm = current->mm;
  91         struct vm_area_struct * vma;
  92         unsigned long task_size = TASK_SIZE;
  93         int do_color_align;
  94         struct vm_unmapped_area_info info;
  95
  96         if (flags & MAP_FIXED) {
  97                 /* We do not accept a shared mapping if it would violate
  98                  * cache aliasing constraints.
  99                  */
 100                 if ((flags & MAP_SHARED) &&
 101                     ((addr - (pgoff << PAGE_SHIFT)) & (SHMLBA - 1)))
 102                         return -EINVAL;
 103                 return addr;
 104         }
 105
 106         if (test_thread_flag(TIF_32BIT))
 107                 task_size = STACK_TOP32;
 108         if (unlikely(len > task_size || len >= VA_EXCLUDE_START))
 109                 return -ENOMEM;
 110
 111         do_color_align = 0;
 112         if (filp || (flags & MAP_SHARED))
 113                 do_color_align = 1;
 114
 115         if (addr) {
 116                 if (do_color_align)
 117                         addr = COLOR_ALIGN(addr, pgoff);
 118                 else
 119                         addr = PAGE_ALIGN(addr);
 120
 121                 vma = find_vma(mm, addr);
 122                 if (task_size - len >= addr &&
 123                     (!vma || addr + len <= vm_start_gap(vma)))
 124                         return addr;
 125         }
 126
 127         info.flags = 0;
 128         info.length = len;
 129         info.low_limit = TASK_UNMAPPED_BASE;
 130         info.high_limit = min(task_size, VA_EXCLUDE_START);
 131         info.align_mask = do_color_align ? (PAGE_MASK & (SHMLBA - 1)) : 0;
 132         info.align_offset = pgoff << PAGE_SHIFT;
 133         addr = vm_unmapped_area(&info);
 134
 135         if ((addr & ~PAGE_MASK) && task_size > VA_EXCLUDE_END) {
 136                 VM_BUG_ON(addr != -ENOMEM);
 137                 info.low_limit = VA_EXCLUDE_END;
 138                 info.high_limit = task_size;
 139                 addr = vm_unmapped_area(&info);
 140         }
 141
 142         return addr;
 143 }
 144
 145 unsigned long
 146 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
 147                           const unsigned long len, const unsigned long pgoff,
 148                           const unsigned long flags)
 149 {
 150         struct vm_area_struct *vma;
 151         struct mm_struct *mm = current->mm;
 152         unsigned long task_size = STACK_TOP32;
 153         unsigned long addr = addr0;
 154         int do_color_align;
 155         struct vm_unmapped_area_info info;
 156
 157         /* This should only ever run for 32-bit processes.  */
 158         BUG_ON(!test_thread_flag(TIF_32BIT));
 159
 160         if (flags & MAP_FIXED) {
 161                 /* We do not accept a shared mapping if it would violate
 162                  * cache aliasing constraints.
 163                  */
 164                 if ((flags & MAP_SHARED) &&
 165                     ((addr - (pgoff << PAGE_SHIFT)) & (SHMLBA - 1)))
 166                         return -EINVAL;
 167                 return addr;
 168         }
 169
 170         if (unlikely(len > task_size))
 171                 return -ENOMEM;
 172
 173         do_color_align = 0;
 174         if (filp || (flags & MAP_SHARED))
 175                 do_color_align = 1;
 176
 177         /* requesting a specific address */
 178         if (addr) {
 179                 if (do_color_align)
 180                         addr = COLOR_ALIGN(addr, pgoff);
 181                 else
 182                         addr = PAGE_ALIGN(addr);
 183
 184                 vma = find_vma(mm, addr);
 185                 if (task_size - len >= addr &&
 186                     (!vma || addr + len <= vm_start_gap(vma)))
 187                         return addr;
 188         }
 189
 190         info.flags = VM_UNMAPPED_AREA_TOPDOWN;
 191         info.length = len;
 192         info.low_limit = PAGE_SIZE;
 193         info.high_limit = mm->mmap_base;
 194         info.align_mask = do_color_align ? (PAGE_MASK & (SHMLBA - 1)) : 0;
 195         info.align_offset = pgoff << PAGE_SHIFT;
 196         addr = vm_unmapped_area(&info);
 197
 198         /*
 199          * A failed mmap() very likely causes application failure,
 200          * so fall back to the bottom-up function here. This scenario
 201          * can happen with large stack limits and large mmap()
 202          * allocations.
 203          */
 204         if (addr & ~PAGE_MASK) {
 205                 VM_BUG_ON(addr != -ENOMEM);
 206                 info.flags = 0;
 207                 info.low_limit = TASK_UNMAPPED_BASE;
 208                 info.high_limit = STACK_TOP32;
 209                 addr = vm_unmapped_area(&info);
 210         }
 211
 212         return addr;
 213 }
 214
 215 /* Try to align mapping such that we align it as much as possible. */
 216 unsigned long get_fb_unmapped_area(struct file *filp, unsigned long orig_addr, unsigned long len, unsigned long pgoff, unsigned long flags)
 217 {
 218         unsigned long align_goal, addr = -ENOMEM;
 219         unsigned long (*get_area)(struct file *, unsigned long,
 220                                   unsigned long, unsigned long, unsigned long);
 221
 222         get_area = current->mm->get_unmapped_area;
 223
 224         if (flags & MAP_FIXED) {
 225                 /* Ok, don't mess with it. */
 226                 return get_area(NULL, orig_addr, len, pgoff, flags);
 227         }
 228         flags &= ~MAP_SHARED;
 229
 230         align_goal = PAGE_SIZE;
 231         if (len >= (4UL * 1024 * 1024))
 232                 align_goal = (4UL * 1024 * 1024);
 233         else if (len >= (512UL * 1024))
 234                 align_goal = (512UL * 1024);
 235         else if (len >= (64UL * 1024))
 236                 align_goal = (64UL * 1024);
 237
 238         do {
 239                 addr = get_area(NULL, orig_addr, len + (align_goal - PAGE_SIZE), pgoff, flags);
 240                 if (!(addr & ~PAGE_MASK)) {
 241                         addr = (addr + (align_goal - 1UL)) & ~(align_goal - 1UL);
 242                         break;
 243                 }
 244
 245                 if (align_goal == (4UL * 1024 * 1024))
 246                         align_goal = (512UL * 1024);
 247                 else if (align_goal == (512UL * 1024))
 248                         align_goal = (64UL * 1024);
 249                 else
 250                         align_goal = PAGE_SIZE;
 251         } while ((addr & ~PAGE_MASK) && align_goal > PAGE_SIZE);
 252
 253         /* Mapping is smaller than 64K or larger areas could not
 254          * be obtained.
 255          */
 256         if (addr & ~PAGE_MASK)
 257                 addr = get_area(NULL, orig_addr, len, pgoff, flags);
 258
 259         return addr;
 260 }
 261 EXPORT_SYMBOL(get_fb_unmapped_area);
 262
 263 /* Essentially the same as PowerPC.  */
 264 static unsigned long mmap_rnd(void)
 265 {
 266         unsigned long rnd = 0UL;
 267
 268         if (current->flags & PF_RANDOMIZE) {
 269                 unsigned long val = get_random_long();
 270                 if (test_thread_flag(TIF_32BIT))
 271                         rnd = (val % (1UL << (23UL-PAGE_SHIFT)));
 272                 else
 273                         rnd = (val % (1UL << (30UL-PAGE_SHIFT)));
 274         }
 275         return rnd << PAGE_SHIFT;
 276 }
 277
 278 void arch_pick_mmap_layout(struct mm_struct *mm)
 279 {
 280         unsigned long random_factor = mmap_rnd();
 281         unsigned long gap;
 282
 283         /*
 284          * Fall back to the standard layout if the personality
 285          * bit is set, or if the expected stack growth is unlimited:
 286          */
 287         gap = rlimit(RLIMIT_STACK);
 288         if (!test_thread_flag(TIF_32BIT) ||
 289             (current->personality & ADDR_COMPAT_LAYOUT) ||
 290             gap == RLIM_INFINITY ||
 291             sysctl_legacy_va_layout) {
 292                 mm->mmap_base = TASK_UNMAPPED_BASE + random_factor;
 293                 mm->get_unmapped_area = arch_get_unmapped_area;
 294         } else {
 295                 /* We know it's 32-bit */
 296                 unsigned long task_size = STACK_TOP32;
 297
 298                 if (gap < 128 * 1024 * 1024)
 299                         gap = 128 * 1024 * 1024;
 300                 if (gap > (task_size / 6 * 5))
 301                         gap = (task_size / 6 * 5);
 302
 303                 mm->mmap_base = PAGE_ALIGN(task_size - gap - random_factor);
 304                 mm->get_unmapped_area = arch_get_unmapped_area_topdown;
 305         }
 306 }
 307
 308 /*
 309  * sys_pipe() is the normal C calling standard for creating
 310  * a pipe. It's not the way unix traditionally does this, though.
 311  */
 312 SYSCALL_DEFINE1(sparc_pipe_real, struct pt_regs *, regs)
 313 {
 314         int fd[2];
 315         int error;
 316
 317         error = do_pipe_flags(fd, 0);
 318         if (error)
 319                 goto out;
 320         regs->u_regs[UREG_I1] = fd[1];
 321         error = fd[0];
 322 out:
 323         return error;
 324 }
 325
 326 /*
 327  * sys_ipc() is the de-multiplexer for the SysV IPC calls..
 328  *
 329  * This is really horribly ugly.
 330  */
 331
 332 SYSCALL_DEFINE6(sparc_ipc, unsigned int, call, int, first, unsigned long, second,
 333                 unsigned long, third, void __user *, ptr, long, fifth)
 334 {
 335         long err;
 336
 337         /* No need for backward compatibility. We can start fresh... */
 338         if (call <= SEMTIMEDOP) {
 339                 switch (call) {
 340                 case SEMOP:
 341                         err = sys_semtimedop(first, ptr,
 342                                              (unsigned int)second, NULL);
 343                         goto out;
 344                 case SEMTIMEDOP:
 345                         err = sys_semtimedop(first, ptr, (unsigned int)second,
 346                                 (const struct timespec __user *)
 347                                              (unsigned long) fifth);
 348                         goto out;
 349                 case SEMGET:
 350                         err = sys_semget(first, (int)second, (int)third);
 351                         goto out;
 352                 case SEMCTL: {
 353                         err = sys_semctl(first, second,
 354                                          (int)third | IPC_64,
 355                                          (unsigned long) ptr);
 356                         goto out;
 357                 }
 358                 default:
 359                         err = -ENOSYS;
 360                         goto out;
 361                 }
 362         }
 363         if (call <= MSGCTL) {
 364                 switch (call) {
 365                 case MSGSND:
 366                         err = sys_msgsnd(first, ptr, (size_t)second,
 367                                          (int)third);
 368                         goto out;
 369                 case MSGRCV:
 370                         err = sys_msgrcv(first, ptr, (size_t)second, fifth,
 371                                          (int)third);
 372                         goto out;
 373                 case MSGGET:
 374                         err = sys_msgget((key_t)first, (int)second);
 375                         goto out;
 376                 case MSGCTL:
 377                         err = sys_msgctl(first, (int)second | IPC_64, ptr);
 378                         goto out;
 379                 default:
 380                         err = -ENOSYS;
 381                         goto out;
 382                 }
 383         }
 384         if (call <= SHMCTL) {
 385                 switch (call) {
 386                 case SHMAT: {
 387                         ulong raddr;
 388                         err = do_shmat(first, ptr, (int)second, &raddr, SHMLBA);
 389                         if (!err) {
 390                                 if (put_user(raddr,
 391                                              (ulong __user *) third))
 392                                         err = -EFAULT;
 393                         }
 394                         goto out;
 395                 }
 396                 case SHMDT:
 397                         err = sys_shmdt(ptr);
 398                         goto out;
 399                 case SHMGET:
 400                         err = sys_shmget(first, (size_t)second, (int)third);
 401                         goto out;
 402                 case SHMCTL:
 403                         err = sys_shmctl(first, (int)second | IPC_64, ptr);
 404                         goto out;
 405                 default:
 406                         err = -ENOSYS;
 407                         goto out;
 408                 }
 409         } else {
 410                 err = -ENOSYS;
 411         }
 412 out:
 413         return err;
 414 }
 415
 416 SYSCALL_DEFINE1(sparc64_personality, unsigned long, personality)
 417 {
 418         long ret;
 419
 420         if (personality(current->personality) == PER_LINUX32 &&
 421             personality(personality) == PER_LINUX)
 422                 personality |= PER_LINUX32;
 423         ret = sys_personality(personality);
 424         if (personality(ret) == PER_LINUX32)
 425                 ret &= ~PER_LINUX32;
 426
 427         return ret;
 428 }
 429
 430 int sparc_mmap_check(unsigned long addr, unsigned long len)
 431 {
 432         if (test_thread_flag(TIF_32BIT)) {
 433                 if (len >= STACK_TOP32)
 434                         return -EINVAL;
 435
 436                 if (addr > STACK_TOP32 - len)
 437                         return -EINVAL;
 438         } else {
 439                 if (len >= VA_EXCLUDE_START)
 440                         return -EINVAL;
 441
 442                 if (invalid_64bit_range(addr, len))
 443                         return -EINVAL;
 444         }
 445
 446         return 0;
 447 }
 448
 449 /* Linux version of mmap */
 450 SYSCALL_DEFINE6(mmap, unsigned long, addr, unsigned long, len,
 451                 unsigned long, prot, unsigned long, flags, unsigned long, fd,
 452                 unsigned long, off)
 453 {
 454         unsigned long retval = -EINVAL;
 455
 456         if ((off + PAGE_ALIGN(len)) < off)
 457                 goto out;
 458         if (off & ~PAGE_MASK)
 459                 goto out;
 460         retval = sys_mmap_pgoff(addr, len, prot, flags, fd, off >> PAGE_SHIFT);
 461 out:
 462         return retval;
 463 }
 464
 465 SYSCALL_DEFINE2(64_munmap, unsigned long, addr, size_t, len)
 466 {
 467         if (invalid_64bit_range(addr, len))
 468                 return -EINVAL;
 469
 470         return vm_munmap(addr, len);
 471 }
 472
 473 SYSCALL_DEFINE5(64_mremap, unsigned long, addr, unsigned long, old_len,
 474                 unsigned long, new_len, unsigned long, flags,
 475                 unsigned long, new_addr)
 476 {
 477         if (test_thread_flag(TIF_32BIT))
 478                 return -EINVAL;
 479         return sys_mremap(addr, old_len, new_len, flags, new_addr);
 480 }
 481
 482 /* we come to here via sys_nis_syscall so it can setup the regs argument */
 483 asmlinkage unsigned long c_sys_nis_syscall(struct pt_regs *regs)
 484 {
 485         static int count;
 486
 487         /* Don't make the system unusable, if someone goes stuck */
 488         if (count++ > 5)
 489                 return -ENOSYS;
 490
 491         printk ("Unimplemented SPARC system call %ld\n",regs->u_regs[1]);
 492 #ifdef DEBUG_UNIMP_SYSCALL
 493         show_regs (regs);
 494 #endif
 495
 496         return -ENOSYS;
 497 }
 498
 499 /* #define DEBUG_SPARC_BREAKPOINT */
 500
 501 asmlinkage void sparc_breakpoint(struct pt_regs *regs)
 502 {
 503         enum ctx_state prev_state = exception_enter();
 504         siginfo_t info;
 505
 506         if (test_thread_flag(TIF_32BIT)) {
 507                 regs->tpc &= 0xffffffff;
 508                 regs->tnpc &= 0xffffffff;
 509         }
 510 #ifdef DEBUG_SPARC_BREAKPOINT
 511         printk ("TRAP: Entering kernel PC=%lx, nPC=%lx\n", regs->tpc, regs->tnpc);
 512 #endif
 513         info.si_signo = SIGTRAP;
 514         info.si_errno = 0;
 515         info.si_code = TRAP_BRKPT;
 516         info.si_addr = (void __user *)regs->tpc;
 517         info.si_trapno = 0;
 518         force_sig_info(SIGTRAP, &info, current);
 519 #ifdef DEBUG_SPARC_BREAKPOINT
 520         printk ("TRAP: Returning to space: PC=%lx nPC=%lx\n", regs->tpc, regs->tnpc);
 521 #endif
 522         exception_exit(prev_state);
 523 }
 524
 525 extern void check_pending(int signum);
 526
 527 SYSCALL_DEFINE2(getdomainname, char __user *, name, int, len)
 528 {
 529         int nlen, err;
 530
 531         if (len < 0)
 532                 return -EINVAL;
 533
 534         down_read(&uts_sem);
 535
 536         nlen = strlen(utsname()->domainname) + 1;
 537         err = -EINVAL;
 538         if (nlen > len)
 539                 goto out;
 540
 541         err = -EFAULT;
 542         if (!copy_to_user(name, utsname()->domainname, nlen))
 543                 err = 0;
 544
 545 out:
 546         up_read(&uts_sem);
 547         return err;
 548 }
 549
 550 SYSCALL_DEFINE5(utrap_install, utrap_entry_t, type,
 551                 utrap_handler_t, new_p, utrap_handler_t, new_d,
 552                 utrap_handler_t __user *, old_p,
 553                 utrap_handler_t __user *, old_d)
 554 {
 555         if (type < UT_INSTRUCTION_EXCEPTION || type > UT_TRAP_INSTRUCTION_31)
 556                 return -EINVAL;
 557         if (new_p == (utrap_handler_t)(long)UTH_NOCHANGE) {
 558                 if (old_p) {
 559                         if (!current_thread_info()->utraps) {
 560                                 if (put_user(NULL, old_p))
 561                                         return -EFAULT;
 562                         } else {
 563                                 if (put_user((utrap_handler_t)(current_thread_info()->utraps[type]), old_p))
 564                                         return -EFAULT;
 565                         }
 566                 }
 567                 if (old_d) {
 568                         if (put_user(NULL, old_d))
 569                                 return -EFAULT;
 570                 }
 571                 return 0;
 572         }
 573         if (!current_thread_info()->utraps) {
 574                 current_thread_info()->utraps =
 575                         kzalloc((UT_TRAP_INSTRUCTION_31+1)*sizeof(long), GFP_KERNEL);
 576                 if (!current_thread_info()->utraps)
 577                         return -ENOMEM;
 578                 current_thread_info()->utraps[0] = 1;
 579         } else {
 580                 if ((utrap_handler_t)current_thread_info()->utraps[type] != new_p &&
 581                     current_thread_info()->utraps[0] > 1) {
 582                         unsigned long *p = current_thread_info()->utraps;
 583
 584                         current_thread_info()->utraps =
 585                                 kmalloc((UT_TRAP_INSTRUCTION_31+1)*sizeof(long),
 586                                         GFP_KERNEL);
 587                         if (!current_thread_info()->utraps) {
 588                                 current_thread_info()->utraps = p;
 589                                 return -ENOMEM;
 590                         }
 591                         p[0]--;
 592                         current_thread_info()->utraps[0] = 1;
 593                         memcpy(current_thread_info()->utraps+1, p+1,
 594                                UT_TRAP_INSTRUCTION_31*sizeof(long));
 595                 }
 596         }
 597         if (old_p) {
 598                 if (put_user((utrap_handler_t)(current_thread_info()->utraps[type]), old_p))
 599                         return -EFAULT;
 600         }
 601         if (old_d) {
 602                 if (put_user(NULL, old_d))
 603                         return -EFAULT;
 604         }
 605         current_thread_info()->utraps[type] = (long)new_p;
 606
 607         return 0;
 608 }
 609
 610 asmlinkage long sparc_memory_ordering(unsigned long model,
 611                                       struct pt_regs *regs)
 612 {
 613         if (model >= 3)
 614                 return -EINVAL;
 615         regs->tstate = (regs->tstate & ~TSTATE_MM) | (model << 14);
 616         return 0;
 617 }
 618
 619 SYSCALL_DEFINE5(rt_sigaction, int, sig, const struct sigaction __user *, act,
 620                 struct sigaction __user *, oact, void __user *, restorer,
 621                 size_t, sigsetsize)
 622 {
 623         struct k_sigaction new_ka, old_ka;
 624         int ret;
 625
 626         /* XXX: Don't preclude handling different sized sigset_t's.  */
 627         if (sigsetsize != sizeof(sigset_t))
 628                 return -EINVAL;
 629
 630         if (act) {
 631                 new_ka.ka_restorer = restorer;
 632                 if (copy_from_user(&new_ka.sa, act, sizeof(*act)))
 633                         return -EFAULT;
 634         }
 635
 636         ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
 637
 638         if (!ret && oact) {
 639                 if (copy_to_user(oact, &old_ka.sa, sizeof(*oact)))
 640                         return -EFAULT;
 641         }
 642
 643         return ret;
 644 }
 645
 646 asmlinkage long sys_kern_features(void)
 647 {
 648         return KERN_FEATURE_MIXED_MODE_STACK;
 649 }